Boston, Mass. A computational physicist and a cognitive neuroscientist at Children's Hospital Boston have come up with the beginnings of a noninvasive test to evaluate an infant's autism risk. It combines the standard electroencephalogram (EEG), which records electrical activity in the brain, with machine-learning algorithms. In a pilot study, their system had 80 percent accuracy in distinguishing between 9-month-old infants known to be at high risk for autism from controls of the same age.
Although this work, published February 22 in the online open-access journal BMC Medicine, requires validation and refinement, it suggests a safe, practical way of identifying infants at high risk for developing autism by capturing very early differences in brain organization and function. This would allow parents to begin behavioral interventions one to two years before autism can be diagnosed through traditional behavioral testing.
"Electrical activity produced by the brain has a lot more information than we realized," says William Bosl, PhD, a neuroinformatics researcher in the Children's Hospital Informatics Program. "Computer algorithms can pick out patterns in those squiggly lines that the eye can't see."
Bosl, Charles A. Nelson, PhD, Research Director of the Developmental Medicine Center at Children's, and colleagues recorded resting EEG signals from 79 babies 6 to 24 months of age participating in a larger study aimed at finding very early risk markers of autism. Forty-six infants had an older sibling with a confirmed diagnosis of an autism spectrum disorder (ASD); the other 33 had no family history of ASDs.
As the babies watched a research assistant blowing bubbles, recordings were made via a hairnet-like cap on their scalps, studded with 64 electrodes. When possible, tests were repeated at 6, 9, 12, 18 and 24 months of age.
Bosl then took the EEG brain-wave readings for each electrode and computed their modified m
|Contact: Keri Stedman|
Children's Hospital Boston